Show code
pacman::p_load(arrow, lubridate, sf, tidyverse, spNetwork, tmap,
spatstat, raster, maptools)Take-home_Ex01
1. Objective
We will be applying appropriate spatial point patterns analysis methods learned in class to discover the geographical and spatio-temporal distribution of Grab hailing services locations in Singapore.
2. Getting Started
2.1 Loading R packages
The R packages that we will be using in this exercise are as follows:
arrow: For reading parquet files (Grab-Posisi Dataset)
lubridate: To handle the date formatting
sf: Import, manage and process vector-based geospatial data in R.
tidyverse: a collection of packages for data science tasks
spatstat: Wide range of useful functions for point pattern analysis and derive kernel density estimation (KDE) layer.
spNetwork: provides functions to perform Spatial Point Patterns Analysis such as kernel density estimation (KDE) and K-function on network. It also can be used to build spatial matrices (‘listw’ objects like in ‘spdep’ package) to conduct any kind of traditional spatial analysis with spatial weights based on reticular distances.
tmap: Provides functions for plotting cartographic quality static point patterns maps or interactive maps by using leaflet API.
raster: reads, writes, manipulates, analyses and model of gridded spatial data (i.e. raster). In this hands-on exercise, it will be used to convert image output generate by spatstat into raster format.
maptools: Provides a set of tools for manipulating geographic data. In this take-home exercise, we mainly use it to convert Spatial objects into ppp format of spatstat.
# classInt, viridis, rgdal
2.2 Importing the datasets
The datasets that we will be using are as follow:
Using read_parquet() function from arrow package to import the grab data, then changing pingtimestamp column to datetime object
Show code
grab_df0 <- read_parquet("data/aspatial/part-00000.snappy.parquet")
grab_df1 <- read_parquet("data/aspatial/part-00001.snappy.parquet")
grab_df2 <- read_parquet("data/aspatial/part-00002.snappy.parquet")
grab_df3 <- read_parquet("data/aspatial/part-00003.snappy.parquet")
grab_df4 <- read_parquet("data/aspatial/part-00004.snappy.parquet")
grab_df5 <- read_parquet("data/aspatial/part-00005.snappy.parquet")
grab_df6 <- read_parquet("data/aspatial/part-00006.snappy.parquet")
grab_df7 <- read_parquet("data/aspatial/part-00007.snappy.parquet")
grab_df8 <- read_parquet("data/aspatial/part-00008.snappy.parquet")
grab_df9 <- read_parquet("data/aspatial/part-00009.snappy.parquet")Show code
grab_df <- bind_rows(grab_df0,
grab_df1,
grab_df2,
grab_df3,
grab_df4,
grab_df5,
grab_df6,
grab_df7,
grab_df8,
grab_df9)
grab_df$pingtimestamp <- as_datetime(grab_df$pingtimestamp)Transforming the coordinate system at the same time when we are importing the data
Show code
sg_roads <- st_read(dsn = "data/geospatial", layer = "gis_osm_roads_free_1") %>% st_transform(crs = 3414)Reading layer `gis_osm_roads_free_1' from data source
`/Users/jacksontan/Documents/Sashimii0219/IS415-GAA/Take-home_Ex/Take-home_Ex01/data/geospatial'
using driver `ESRI Shapefile'
Simple feature collection with 1759836 features and 10 fields
Geometry type: LINESTRING
Dimension: XY
Bounding box: xmin: 99.66041 ymin: 0.8021131 xmax: 119.2601 ymax: 7.514393
Geodetic CRS: WGS 84Transforming the coordinate system at the same time when we are importing the data
Show code
mpsz2019 <- st_read("data/geospatial", layer = "MPSZ-2019") %>% st_transform(crs = 3414)Reading layer `MPSZ-2019' from data source
`/Users/jacksontan/Documents/Sashimii0219/IS415-GAA/Take-home_Ex/Take-home_Ex01/data/geospatial'
using driver `ESRI Shapefile'
Simple feature collection with 332 features and 6 fields
Geometry type: MULTIPOLYGON
Dimension: XY
Bounding box: xmin: 103.6057 ymin: 1.158699 xmax: 104.0885 ymax: 1.470775
Geodetic CRS: WGS 843. Geospatial Data Wrangling
Before we begin exploring the data, we will first need to perform some data pre-processing on the datasets that we have imported.
3.1 Data Pre-processing - MPSZ2019
3.1.1 Excluding Outer Islands
As grab won’t be able to reach offshore places, we will exclude the outer islands from this dataset. We will do this through the following steps:
We will first take a look at the unique planning areas in Singapore using unique() on the PLN_AREA_N column of mpsz2019 dataset.
Show code
unique(mpsz2019$PLN_AREA_N) [1] "MARINA EAST" "RIVER VALLEY"
[3] "SINGAPORE RIVER" "WESTERN ISLANDS"
[5] "MUSEUM" "MARINE PARADE"
[7] "SOUTHERN ISLANDS" "BUKIT MERAH"
[9] "DOWNTOWN CORE" "STRAITS VIEW"
[11] "QUEENSTOWN" "OUTRAM"
[13] "MARINA SOUTH" "ROCHOR"
[15] "KALLANG" "TANGLIN"
[17] "NEWTON" "CLEMENTI"
[19] "BEDOK" "PIONEER"
[21] "JURONG EAST" "ORCHARD"
[23] "GEYLANG" "BOON LAY"
[25] "BUKIT TIMAH" "NOVENA"
[27] "TOA PAYOH" "TUAS"
[29] "JURONG WEST" "SERANGOON"
[31] "BISHAN" "TAMPINES"
[33] "BUKIT BATOK" "HOUGANG"
[35] "CHANGI BAY" "PAYA LEBAR"
[37] "ANG MO KIO" "PASIR RIS"
[39] "BUKIT PANJANG" "TENGAH"
[41] "SELETAR" "SUNGEI KADUT"
[43] "YISHUN" "MANDAI"
[45] "PUNGGOL" "CHOA CHU KANG"
[47] "SENGKANG" "CHANGI"
[49] "CENTRAL WATER CATCHMENT" "SEMBAWANG"
[51] "WESTERN WATER CATCHMENT" "WOODLANDS"
[53] "NORTH-EASTERN ISLANDS" "SIMPANG"
[55] "LIM CHU KANG" Show code
plot(mpsz2019)
Note that there are 3 areas with island in their name, mainly “NORTH-EASTERN ISLANDS”, “SOUTHERN ISLANDS”, and “WESTERN ISLANDS”.
To exclude the islands, we simply have to pass a condition to exclude these islands in the subset function.
Show code
mpsz2019_new <- subset(mpsz2019, !(PLN_AREA_N %in%
c("NORTH-EASTERN ISLANDS", "SOUTHERN ISLANDS", "WESTERN ISLANDS")))Great! Now let’s check if we indeed removed the maps!
Show code
tmap_mode('plot')
before <- tm_shape(mpsz2019) +
tm_polygons("PLN_AREA_N") +
tmap_options(max.categories = 53)
after <- tm_shape(mpsz2019_new) +
tm_polygons("PLN_AREA_N") +
tmap_options(max.categories = 53)
tmap_arrange(before, after)
3.1.2 Invalid Geometries
We will be using the st_is_valid() function to test for invalid geometries.
Show code
test <- st_is_valid(mpsz2019_new,reason=TRUE)
# Number of invalid geometries
length(which(test!= "Valid Geometry"))[1] 3Show code
# Reason
test[which(test!= "Valid Geometry")][1] "Ring Self-intersection[26922.5243000389 27027.610899987]"
[2] "Ring Self-intersection[38991.2589000446 31986.5599999869]"
[3] "Ring Self-intersection[14484.6860000313 31330.1319999856]"We can see that there are 3 invalid geometries. Let’s fix them using st_make_valid().
Show code
mpsz2019_new<- st_make_valid(mpsz2019_new)
length(which(st_is_valid(mpsz2019_new) == FALSE))[1] 03.1.3 Missing Values
Show code
mpsz2019_new[rowSums(is.na(mpsz2019_new))!=0,]Simple feature collection with 0 features and 6 fields
Bounding box: xmin: NA ymin: NA xmax: NA ymax: NA
Projected CRS: SVY21 / Singapore TM
[1] SUBZONE_N SUBZONE_C PLN_AREA_N PLN_AREA_C REGION_N REGION_C geometry
<0 rows> (or 0-length row.names)Using the code above, we can see that there are no missing values.
3.1.4 Creating boundary?
Show code
sg_boundary <- mpsz2019_new %>% st_union()
plot(sg_boundary)
3.2 Data Pre-processing - OpenStreetMap Road Dataset
3.2.1 Limiting the dataset
As the dataset contains data from Malaysia and Brunei as well, we will use st_intersection() to limit the data to only Singapore.
Show code
points_within_sg <- st_intersection(sg_roads, mpsz2019_new)Now, we can see that in points_within_sg it only contain Singapore road data, combined with the other values from mpsz2019 like “PLN_AREA_N” used above.
Show code
colnames(points_within_sg) [1] "osm_id" "code" "fclass" "name" "ref"
[6] "oneway" "maxspeed" "layer" "bridge" "tunnel"
[11] "SUBZONE_N" "SUBZONE_C" "PLN_AREA_N" "PLN_AREA_C" "REGION_N"
[16] "REGION_C" "geometry" Show code
head(points_within_sg)Simple feature collection with 6 features and 16 fields
Geometry type: LINESTRING
Dimension: XY
Bounding box: xmin: 31466.72 ymin: 30680.54 xmax: 32815.21 ymax: 30873.74
Projected CRS: SVY21 / Singapore TM
osm_id code fclass name ref oneway maxspeed layer
4052 23946437 5122 residential Rhu Cross <NA> F 50 0
9668 32605139 5131 motorway_link <NA> <NA> F 40 0
20076 46337834 5131 motorway_link <NA> <NA> F 50 -2
21690 49961799 5111 motorway East Coast Parkway ECP F 70 1
26543 74722808 5111 motorway East Coast Parkway ECP F 70 1
29808 99007260 5131 motorway_link <NA> <NA> F 50 1
bridge tunnel SUBZONE_N SUBZONE_C PLN_AREA_N PLN_AREA_C REGION_N
4052 F F MARINA EAST MESZ01 MARINA EAST ME CENTRAL REGION
9668 F F MARINA EAST MESZ01 MARINA EAST ME CENTRAL REGION
20076 F T MARINA EAST MESZ01 MARINA EAST ME CENTRAL REGION
21690 F F MARINA EAST MESZ01 MARINA EAST ME CENTRAL REGION
26543 T F MARINA EAST MESZ01 MARINA EAST ME CENTRAL REGION
29808 T F MARINA EAST MESZ01 MARINA EAST ME CENTRAL REGION
REGION_C geometry
4052 CR LINESTRING (31889.45 30760....
9668 CR LINESTRING (32768.57 30857....
20076 CR LINESTRING (32815.21 30873....
21690 CR LINESTRING (32365.45 30845....
26543 CR LINESTRING (31611.63 30720....
29808 CR LINESTRING (31611.63 30720....3.2.2 Invalid Geometries
Again, using the st_is_valid() function to test for invalid geometries.
Show code
test <- st_is_valid(points_within_sg,reason=TRUE)
# Number of invalid geometries
length(which(test!= "Valid Geometry"))[1] 0Show code
# Reason
test[which(test!= "Valid Geometry")]character(0)No invalid geometries!
3.2.3 Missing Values / Dropping Columns
Show code
points_within_sg[rowSums(is.na(points_within_sg))!=0,]Simple feature collection with 232766 features and 16 fields
Geometry type: GEOMETRY
Dimension: XY
Bounding box: xmin: 2679.373 ymin: 23099.51 xmax: 50957.8 ymax: 50220.06
Projected CRS: SVY21 / Singapore TM
First 10 features:
osm_id code fclass name ref oneway maxspeed layer
4052 23946437 5122 residential Rhu Cross <NA> F 50 0
9668 32605139 5131 motorway_link <NA> <NA> F 40 0
20076 46337834 5131 motorway_link <NA> <NA> F 50 -2
29808 99007260 5131 motorway_link <NA> <NA> F 50 1
45723 140562813 5131 motorway_link <NA> <NA> F 70 -1
45728 140562819 5131 motorway_link <NA> <NA> F 50 0
45731 140562823 5131 motorway_link <NA> <NA> F 60 -2
45733 140562826 5131 motorway_link <NA> <NA> F 40 0
52966 150819034 5141 service Bay East Drive <NA> B 0 0
84664 174717984 5153 footway <NA> <NA> B 0 0
bridge tunnel SUBZONE_N SUBZONE_C PLN_AREA_N PLN_AREA_C REGION_N
4052 F F MARINA EAST MESZ01 MARINA EAST ME CENTRAL REGION
9668 F F MARINA EAST MESZ01 MARINA EAST ME CENTRAL REGION
20076 F T MARINA EAST MESZ01 MARINA EAST ME CENTRAL REGION
29808 T F MARINA EAST MESZ01 MARINA EAST ME CENTRAL REGION
45723 F F MARINA EAST MESZ01 MARINA EAST ME CENTRAL REGION
45728 F F MARINA EAST MESZ01 MARINA EAST ME CENTRAL REGION
45731 F T MARINA EAST MESZ01 MARINA EAST ME CENTRAL REGION
45733 F F MARINA EAST MESZ01 MARINA EAST ME CENTRAL REGION
52966 F F MARINA EAST MESZ01 MARINA EAST ME CENTRAL REGION
84664 F F MARINA EAST MESZ01 MARINA EAST ME CENTRAL REGION
REGION_C geometry
4052 CR LINESTRING (31889.45 30760....
9668 CR LINESTRING (32768.57 30857....
20076 CR LINESTRING (32815.21 30873....
29808 CR LINESTRING (31611.63 30720....
45723 CR LINESTRING (32782.42 30754....
45728 CR LINESTRING (32645.37 30683....
45731 CR LINESTRING (32809.68 30108....
45733 CR LINESTRING (32609.11 30700....
52966 CR LINESTRING (32173.46 30036....
84664 CR LINESTRING (31750.06 30644....By using the code above, we can see that majority of the missing values are in the ‘name’ and ‘ref’ column. Therefore, let’s drop the irrelevant columns first before we try it again!
Show code
sg_roads_new <- points_within_sg[c("osm_id", "code", "fclass", "PLN_AREA_N", "geometry")]We only kept “osm_id”, “code”, “fclass”, and “PLN_AREA_N” columns.
Show code
sg_roads_new[rowSums(is.na(sg_roads_new))!=0,]Simple feature collection with 0 features and 4 fields
Bounding box: xmin: NA ymin: NA xmax: NA ymax: NA
Projected CRS: SVY21 / Singapore TM
[1] osm_id code fclass PLN_AREA_N geometry
<0 rows> (or 0-length row.names)No more missing values here.
Our map so far:
Show code
tm_shape(sg_boundary) +
tm_polygons() +
tm_shape(sg_roads_new) +
tm_lines("PLN_AREA_N")
3.3 Data Pre-processing - Grab-Posisi Dataset
The Grab-Posisi Dataset is an Aspatial dataset, different from the two we prepared above. As such, the pre-processing is slightly different too.
3.3.1 Getting the Origin and Destination Locations
The code below is a chain of dplyr pipes to group the trips by their id and extract the first pingtimestamp row of each trip in order to get the origin of it.
Show code
origin_df <- grab_df %>%
group_by(trj_id) %>%
arrange(pingtimestamp) %>%
filter(row_number()==1) %>%
mutate(weekday = wday(pingtimestamp,
label=TRUE,
abbr=TRUE),
start_hr = factor(hour(pingtimestamp)),
day = factor(mday(pingtimestamp)))Show code
destination_df <- grab_df %>%
group_by(trj_id) %>%
arrange(desc(pingtimestamp)) %>%
# Same as previous code but desc, so ending location
filter(row_number()==1) %>%
mutate(weekday = wday(pingtimestamp,
label=TRUE,
abbr=TRUE),
end_hr = factor(hour(pingtimestamp)),
day = factor(mday(pingtimestamp)))3.3.2 Converting to SF format from Dataframe
We will need the files in SF format first before we can use it for further geospatial analysis.
Show code
origin_sf <- st_as_sf(origin_df,
coords = c("rawlng", "rawlat"),
crs=4326) %>%
st_transform(crs = 3414)
dest_sf <- st_as_sf(destination_df,
coords = c("rawlng", "rawlat"),
crs=4326) %>%
st_transform(crs = 3414)3.3.3 Invalid Geometries
Show code
test <- st_is_valid(origin_sf,reason=TRUE)
length(which(test!= "Valid Geometry"))[1] 0Show code
test <- st_is_valid(dest_sf,reason=TRUE)
length(which(test!= "Valid Geometry"))[1] 03.3.4 Missing Files
Show code
origin_sf[rowSums(is.na(origin_sf))!=0,]Simple feature collection with 0 features and 10 fields
Bounding box: xmin: NA ymin: NA xmax: NA ymax: NA
Projected CRS: SVY21 / Singapore TM
# A tibble: 0 × 11
# Groups: trj_id [0]
# ℹ 11 variables: trj_id <chr>, driving_mode <chr>, osname <chr>,
# pingtimestamp <dttm>, speed <dbl>, bearing <int>, accuracy <dbl>,
# weekday <ord>, start_hr <fct>, day <fct>, geometry <GEOMETRY [m]>Show code
dest_sf[rowSums(is.na(dest_sf))!=0,]Simple feature collection with 0 features and 10 fields
Bounding box: xmin: NA ymin: NA xmax: NA ymax: NA
Projected CRS: SVY21 / Singapore TM
# A tibble: 0 × 11
# Groups: trj_id [0]
# ℹ 11 variables: trj_id <chr>, driving_mode <chr>, osname <chr>,
# pingtimestamp <dttm>, speed <dbl>, bearing <int>, accuracy <dbl>,
# weekday <ord>, end_hr <fct>, day <fct>, geometry <GEOMETRY [m]>No missing values, we are almost ready.
3.3.5 Removing points on the islands
Show code
origin_sf_new <- st_intersection(origin_sf, mpsz2019_new)
dest_sf_new <- st_intersection(dest_sf, mpsz2019_new)To verify that the points that we removed is indeed from the islands, here’s a chunk of code to prove:
Show code
# Finding out points removed
diff_id <- origin_sf$trj_id[!(origin_sf$trj_id %in% origin_sf_new$trj_id)]
# Extracting full information of these points
outliers <- origin_sf[(origin_sf$trj_id %in% diff_id), ]
# Checking where these places are from
unique(st_intersection(outliers, mpsz2019)$PLN_AREA_N)[1] "WESTERN ISLANDS" "SOUTHERN ISLANDS"They are indeed from “WESTERN ISLANDS” and “SOUTHERN ISLANDS”.
3.3.6 Dropping Unnecessary Columns
Now that our grab dataset is almost ready, we need to decide which column we should drop. Here are the columns in both origin_sf_new and dest_sf_new:
Show code
colnames(origin_sf_new) [1] "trj_id" "driving_mode" "osname" "pingtimestamp"
[5] "speed" "bearing" "accuracy" "weekday"
[9] "start_hr" "day" "SUBZONE_N" "SUBZONE_C"
[13] "PLN_AREA_N" "PLN_AREA_C" "REGION_N" "REGION_C"
[17] "geometry" Show code
colnames(dest_sf_new) [1] "trj_id" "driving_mode" "osname" "pingtimestamp"
[5] "speed" "bearing" "accuracy" "weekday"
[9] "end_hr" "day" "SUBZONE_N" "SUBZONE_C"
[13] "PLN_AREA_N" "PLN_AREA_C" "REGION_N" "REGION_C"
[17] "geometry" We will definitely be dropping the columns merged from mpsz2019_new (other than PLN_AREA_N), but what about “driving_mode”, “osname”, “speed”, “bearing”, and “accuracy”? Let’s first take a look at them.
Show code
unique(origin_sf_new$driving_mode)[1] "car"Seeing that there is only 1 constant in the column, it is safe for us to drop this column.
Show code
unique(origin_sf_new$osname)[1] "ios" "android"There are 2 values, mainly “ios” and “android”. Arguments can be made that we can analyse the behavior of both type in terms of using grab hailing services, but that’s not what we will doing so we will drop this as well.
As we are analysing start/stop points, speed will not be a relevant factor hence we will be dropping them.
Not relevant as well, hence dropping.
According to research paper published on Grab website, this is the definition of the accuracy column:
“…the accuracy level roughly indicates the radius of the circle within which the true location lies with a certain probability. The lower the accuracy level, the more precise the reported GPS ping is.”
With that, let’s take a look at the distribution of accuracy score.
Show code
plot(origin_sf_new$accuracy)
Show code
ggplot(origin_sf_new,
aes(x=rownames(origin_sf_new), y=accuracy)) +
geom_point(size = 2)
Show code
ggplot(dest_sf_new,
aes(x=rownames(dest_sf_new), y=accuracy)) +
geom_point(size = 2)
From the plot, we can see that there are 3 clear outliers with accuracy above 180~ for origin_sf_new, and 1 for dest_sf_new. Now let’s extract these trips.
Show code
origin_sf_new[origin_sf_new$accuracy > 180, ]Simple feature collection with 1 feature and 16 fields
Geometry type: POINT
Dimension: XY
Bounding box: xmin: 29008.44 ymin: 32353.78 xmax: 29008.44 ymax: 32353.78
Projected CRS: SVY21 / Singapore TM
# A tibble: 1 × 17
trj_id driving_mode osname pingtimestamp speed bearing accuracy weekday
<chr> <chr> <chr> <dttm> <dbl> <int> <dbl> <ord>
1 59560 car ios 2019-04-16 00:28:59 -1 26 728 Tue
# ℹ 9 more variables: start_hr <fct>, day <fct>, SUBZONE_N <chr>,
# SUBZONE_C <chr>, PLN_AREA_N <chr>, PLN_AREA_C <chr>, REGION_N <chr>,
# REGION_C <chr>, geometry <POINT [m]>Show code
dest_sf_new[dest_sf_new$accuracy > 500, ]Simple feature collection with 7 features and 16 fields
Geometry type: POINT
Dimension: XY
Bounding box: xmin: 28983.51 ymin: 29952.1 xmax: 33721.09 ymax: 34502.5
Projected CRS: SVY21 / Singapore TM
# A tibble: 7 × 17
trj_id driving_mode osname pingtimestamp speed bearing accuracy weekday
<chr> <chr> <chr> <dttm> <dbl> <int> <dbl> <ord>
1 54788 car ios 2019-04-09 11:06:56 -1 203 1414 Tue
2 14443 car ios 2019-04-11 01:09:18 -1 223 1414 Thu
3 36434 car ios 2019-04-18 07:41:35 -1 126 806 Thu
4 60701 car ios 2019-04-13 03:16:02 -1 130 818 Sat
5 24103 car ios 2019-04-11 13:19:30 -1 117 1402 Thu
6 58922 car ios 2019-04-13 15:50:44 -1 31 1414 Sat
7 68340 car ios 2019-04-12 11:55:48 -1 10 1414 Fri
# ℹ 9 more variables: end_hr <fct>, day <fct>, SUBZONE_N <chr>,
# SUBZONE_C <chr>, PLN_AREA_N <chr>, PLN_AREA_C <chr>, REGION_N <chr>,
# REGION_C <chr>, geometry <POINT [m]>To ensure that our data is of utmost accuracy, we will drop these trips, before we drop the accuracy column as well (as we will not need it anymore).
Show code
origin_sf_new <- subset(origin_sf_new, accuracy < 180)
dest_sf_new <- subset(dest_sf_new, accuracy < 500)With that done, we can now drop the columns that we don’t need.
Show code
origin_sf_new <- origin_sf_new[, c(1, 4, 8:10, 13, 17)]
dest_sf_new <- dest_sf_new[, c(1, 4, 8:10, 13, 17)]3.3.7 Duplicated Points
Lastly, let’s check for duplicated points on the map.
Show code
# Check for any duplicates
any(duplicated(origin_sf_new))[1] FALSEShow code
# Count the number of duplicates
sum(multiplicity(origin_sf_new) > 1)[1] 0No duplicated points!
3.4 Verifying Coordinate System
It is important for the data to be in the right coordinate reference system (CRS). In this assignment, all spatial data will be projected in EPSG:3414, which is a projected coordinate system for Singapore.
Show code
st_crs(mpsz2019_new)Coordinate Reference System:
User input: EPSG:3414
wkt:
PROJCRS["SVY21 / Singapore TM",
BASEGEOGCRS["SVY21",
DATUM["SVY21",
ELLIPSOID["WGS 84",6378137,298.257223563,
LENGTHUNIT["metre",1]]],
PRIMEM["Greenwich",0,
ANGLEUNIT["degree",0.0174532925199433]],
ID["EPSG",4757]],
CONVERSION["Singapore Transverse Mercator",
METHOD["Transverse Mercator",
ID["EPSG",9807]],
PARAMETER["Latitude of natural origin",1.36666666666667,
ANGLEUNIT["degree",0.0174532925199433],
ID["EPSG",8801]],
PARAMETER["Longitude of natural origin",103.833333333333,
ANGLEUNIT["degree",0.0174532925199433],
ID["EPSG",8802]],
PARAMETER["Scale factor at natural origin",1,
SCALEUNIT["unity",1],
ID["EPSG",8805]],
PARAMETER["False easting",28001.642,
LENGTHUNIT["metre",1],
ID["EPSG",8806]],
PARAMETER["False northing",38744.572,
LENGTHUNIT["metre",1],
ID["EPSG",8807]]],
CS[Cartesian,2],
AXIS["northing (N)",north,
ORDER[1],
LENGTHUNIT["metre",1]],
AXIS["easting (E)",east,
ORDER[2],
LENGTHUNIT["metre",1]],
USAGE[
SCOPE["Cadastre, engineering survey, topographic mapping."],
AREA["Singapore - onshore and offshore."],
BBOX[1.13,103.59,1.47,104.07]],
ID["EPSG",3414]]Show code
st_crs(sg_roads_new)Coordinate Reference System:
User input: EPSG:3414
wkt:
PROJCRS["SVY21 / Singapore TM",
BASEGEOGCRS["SVY21",
DATUM["SVY21",
ELLIPSOID["WGS 84",6378137,298.257223563,
LENGTHUNIT["metre",1]]],
PRIMEM["Greenwich",0,
ANGLEUNIT["degree",0.0174532925199433]],
ID["EPSG",4757]],
CONVERSION["Singapore Transverse Mercator",
METHOD["Transverse Mercator",
ID["EPSG",9807]],
PARAMETER["Latitude of natural origin",1.36666666666667,
ANGLEUNIT["degree",0.0174532925199433],
ID["EPSG",8801]],
PARAMETER["Longitude of natural origin",103.833333333333,
ANGLEUNIT["degree",0.0174532925199433],
ID["EPSG",8802]],
PARAMETER["Scale factor at natural origin",1,
SCALEUNIT["unity",1],
ID["EPSG",8805]],
PARAMETER["False easting",28001.642,
LENGTHUNIT["metre",1],
ID["EPSG",8806]],
PARAMETER["False northing",38744.572,
LENGTHUNIT["metre",1],
ID["EPSG",8807]]],
CS[Cartesian,2],
AXIS["northing (N)",north,
ORDER[1],
LENGTHUNIT["metre",1]],
AXIS["easting (E)",east,
ORDER[2],
LENGTHUNIT["metre",1]],
USAGE[
SCOPE["Cadastre, engineering survey, topographic mapping."],
AREA["Singapore - onshore and offshore."],
BBOX[1.13,103.59,1.47,104.07]],
ID["EPSG",3414]]Show code
st_crs(origin_sf_new)Coordinate Reference System:
User input: EPSG:3414
wkt:
PROJCRS["SVY21 / Singapore TM",
BASEGEOGCRS["SVY21",
DATUM["SVY21",
ELLIPSOID["WGS 84",6378137,298.257223563,
LENGTHUNIT["metre",1]]],
PRIMEM["Greenwich",0,
ANGLEUNIT["degree",0.0174532925199433]],
ID["EPSG",4757]],
CONVERSION["Singapore Transverse Mercator",
METHOD["Transverse Mercator",
ID["EPSG",9807]],
PARAMETER["Latitude of natural origin",1.36666666666667,
ANGLEUNIT["degree",0.0174532925199433],
ID["EPSG",8801]],
PARAMETER["Longitude of natural origin",103.833333333333,
ANGLEUNIT["degree",0.0174532925199433],
ID["EPSG",8802]],
PARAMETER["Scale factor at natural origin",1,
SCALEUNIT["unity",1],
ID["EPSG",8805]],
PARAMETER["False easting",28001.642,
LENGTHUNIT["metre",1],
ID["EPSG",8806]],
PARAMETER["False northing",38744.572,
LENGTHUNIT["metre",1],
ID["EPSG",8807]]],
CS[Cartesian,2],
AXIS["northing (N)",north,
ORDER[1],
LENGTHUNIT["metre",1]],
AXIS["easting (E)",east,
ORDER[2],
LENGTHUNIT["metre",1]],
USAGE[
SCOPE["Cadastre, engineering survey, topographic mapping."],
AREA["Singapore - onshore and offshore."],
BBOX[1.13,103.59,1.47,104.07]],
ID["EPSG",3414]]Show code
st_crs(dest_sf_new)Coordinate Reference System:
User input: EPSG:3414
wkt:
PROJCRS["SVY21 / Singapore TM",
BASEGEOGCRS["SVY21",
DATUM["SVY21",
ELLIPSOID["WGS 84",6378137,298.257223563,
LENGTHUNIT["metre",1]]],
PRIMEM["Greenwich",0,
ANGLEUNIT["degree",0.0174532925199433]],
ID["EPSG",4757]],
CONVERSION["Singapore Transverse Mercator",
METHOD["Transverse Mercator",
ID["EPSG",9807]],
PARAMETER["Latitude of natural origin",1.36666666666667,
ANGLEUNIT["degree",0.0174532925199433],
ID["EPSG",8801]],
PARAMETER["Longitude of natural origin",103.833333333333,
ANGLEUNIT["degree",0.0174532925199433],
ID["EPSG",8802]],
PARAMETER["Scale factor at natural origin",1,
SCALEUNIT["unity",1],
ID["EPSG",8805]],
PARAMETER["False easting",28001.642,
LENGTHUNIT["metre",1],
ID["EPSG",8806]],
PARAMETER["False northing",38744.572,
LENGTHUNIT["metre",1],
ID["EPSG",8807]]],
CS[Cartesian,2],
AXIS["northing (N)",north,
ORDER[1],
LENGTHUNIT["metre",1]],
AXIS["easting (E)",east,
ORDER[2],
LENGTHUNIT["metre",1]],
USAGE[
SCOPE["Cadastre, engineering survey, topographic mapping."],
AREA["Singapore - onshore and offshore."],
BBOX[1.13,103.59,1.47,104.07]],
ID["EPSG",3414]]They are all in the correct CRS!
3.5 Plotting Spatial Data
Finally, plotting all three datasets together to ensure that they have a consistent projection system.
Show code
tm_shape(sg_boundary) +
tm_polygons() +
tm_shape(sg_roads_new) +
tm_lines("PLN_AREA_N") +
tm_shape(origin_sf_new) +
tm_dots()
3.6 Exploratory Data Analysis
Before we begin our Geospatial Analysis, let’s first take a closer look at the Grab dataset.
3.6.1 Day of the Week
The distribution of the trips across all 7 days of the week looks even.
Show code
ggplot(origin_sf_new, aes(x=weekday)) + geom_bar()
Show code
ggplot(dest_sf_new, aes(x=weekday)) + geom_bar()
3.6.2 Planning Area
First let us look at the top 10 planning areas for grab ride origin points. Tampines is the Planning Area with the most origin points.
Show code
origin_pl_area <- origin_sf_new %>%
group_by(PLN_AREA_N) %>%
summarise(total_count=n()) %>%
top_n(10, total_count) %>%
.$PLN_AREA_N
ggplot(origin_sf_new[origin_sf_new$PLN_AREA_N %in% origin_pl_area,],
aes(x=PLN_AREA_N)) + geom_bar() +
theme(axis.text.x = element_text(angle = 90, vjust = 0.5, hjust=1)) +
labs(title = "Trips Origin Distribution by Planning Area",
x = "Planning Area",
y = "Number of Trips")
Then for the destination points.
Show code
dest_pl_area <- dest_sf_new %>%
group_by(PLN_AREA_N) %>%
summarise(total_count=n()) %>%
top_n(10, total_count) %>%
.$PLN_AREA_N
ggplot(dest_sf_new[dest_sf_new$PLN_AREA_N %in% dest_pl_area,],
aes(x=PLN_AREA_N)) + geom_bar() +
theme(axis.text.x = element_text(angle = 90, vjust = 0.5, hjust=1)) +
labs(title = "Trips Destination Distribution by Planning Area",
x = "Planning Area",
y = "Number of Trips")
6 out of 10 of the Planning Areas remains the same for destination points, mainly TAMPINES, WOODLANDS, YISHUN, QUEENSTOWN, BUKIT MERAH, and CHANGI. This time however, the Planning Area with the most destination points is Changi.
3.6.3 Starting Hour
Show code
origin_sf_new$start_hr <- factor(origin_sf_new$start_hr, levels = 0:23)
ggplot(origin_sf_new, aes(x = start_hr)) +
geom_bar() +
labs(title = "Trips Distribution by Start Hour",
x = "Start Hour",
y = "Number of Trips")
From the graph, we can see that the starting hour peaks at midnight (12am - 1am) and morning (9am - 10am), the former probably due to the lack of public transport after operating hours, and the latter from rush hour.
4. Kernel Density Estimation (KDE) Layers
4.1 Converting data format
4.1.1 Creating point ppp objects
In the code chunk below, as.ppp() function is used to derive a ppp object layer directly from a sf tibble data.frame.
Show code
origin_ppp <- as.ppp(origin_sf_new)
summary(origin_ppp)Marked planar point pattern: 27871 points
Average intensity 2.631202e-05 points per square unit
Coordinates are given to 3 decimal places
i.e. rounded to the nearest multiple of 0.001 units
marks are of type 'character'
Summary:
Length Class Mode
27871 character character
Window: rectangle = [3628.24, 49845.23] x [26770.58, 49689.64] units
(46220 x 22920 units)
Window area = 1059250000 square unitsShow code
dest_ppp <- as.ppp(dest_sf_new)
summary(dest_ppp)Marked planar point pattern: 27811 points
Average intensity 2.645533e-05 points per square unit
Coordinates are given to 3 decimal places
i.e. rounded to the nearest multiple of 0.001 units
marks are of type 'character'
Summary:
Length Class Mode
27811 character character
Window: rectangle = [3637.21, 49870.63] x [26770.04, 49507.79] units
(46230 x 22740 units)
Window area = 1051240000 square units4.1.2 Creating owin objects
In the code chunk as.owin() is used to create an owin object class from polygon sf tibble data.frame. In this case, we will be converting the sg_boundary polygon.
Show code
sg_boundary_owin <- as.owin(sg_boundary)4.1.3 Combining point events object and owin object
We will now combine singapore’s boundary and the origin and destination points into one.
Show code
originSG_ppp = origin_ppp[sg_boundary_owin]
destSG_ppp = dest_ppp[sg_boundary_owin]Show code
plot(destSG_ppp)
Show code
summary(destSG_ppp)Marked planar point pattern: 27811 points
Average intensity 4.184642e-05 points per square unit
Coordinates are given to 3 decimal places
i.e. rounded to the nearest multiple of 0.001 units
marks are of type 'character'
Summary:
Length Class Mode
27811 character character
Window: polygonal boundary
37 separate polygons (29 holes)
vertices area relative.area
polygon 1 12666 6.63014e+08 9.98e-01
polygon 2 285 1.61128e+06 2.42e-03
polygon 3 27 1.50315e+04 2.26e-05
polygon 4 (hole) 41 -4.01660e+04 -6.04e-05
polygon 5 (hole) 317 -5.11280e+04 -7.69e-05
polygon 6 (hole) 3 -4.14099e-04 -6.23e-13
polygon 7 30 2.80002e+04 4.21e-05
polygon 8 (hole) 4 -2.86396e-01 -4.31e-10
polygon 9 (hole) 3 -1.81439e-04 -2.73e-13
polygon 10 (hole) 3 -8.68789e-04 -1.31e-12
polygon 11 (hole) 3 -5.99535e-04 -9.02e-13
polygon 12 (hole) 3 -3.04561e-04 -4.58e-13
polygon 13 (hole) 3 -4.46076e-04 -6.71e-13
polygon 14 (hole) 3 -3.39794e-04 -5.11e-13
polygon 15 (hole) 3 -4.52043e-05 -6.80e-14
polygon 16 (hole) 3 -3.90173e-05 -5.87e-14
polygon 17 (hole) 3 -9.59850e-05 -1.44e-13
polygon 18 (hole) 4 -2.54488e-04 -3.83e-13
polygon 19 (hole) 4 -4.28453e-01 -6.45e-10
polygon 20 (hole) 4 -2.18616e-04 -3.29e-13
polygon 21 (hole) 5 -2.44411e-04 -3.68e-13
polygon 22 (hole) 5 -3.64686e-02 -5.49e-11
polygon 23 71 8.18750e+03 1.23e-05
polygon 24 (hole) 6 -8.37554e-01 -1.26e-09
polygon 25 (hole) 38 -7.79904e+03 -1.17e-05
polygon 26 (hole) 3 -3.41897e-05 -5.14e-14
polygon 27 (hole) 3 -3.65499e-03 -5.50e-12
polygon 28 (hole) 3 -4.95057e-02 -7.45e-11
polygon 29 91 1.49663e+04 2.25e-05
polygon 30 (hole) 5 -2.92235e-04 -4.40e-13
polygon 31 (hole) 3 -7.43616e-06 -1.12e-14
polygon 32 (hole) 270 -1.21455e+03 -1.83e-06
polygon 33 (hole) 19 -4.39650e+00 -6.62e-09
polygon 34 (hole) 35 -1.38385e+02 -2.08e-07
polygon 35 (hole) 23 -1.99656e+01 -3.00e-08
polygon 36 71 5.63061e+03 8.47e-06
polygon 37 10 1.99717e+02 3.01e-07
enclosing rectangle: [2667.54, 55941.94] x [21448.47, 50256.33] units
(53270 x 28810 units)
Window area = 664597000 square units
Fraction of frame area: 0.4334.1.4 Rescale
The density values of the output range from 0 to 0.000035 which is way too small to comprehend, and it is computed in “number of points per square meter”. Therefore, we are going to use rescale() to covert the unit of measurement from meter to kilometer.
Show code
originSG_ppp.km <- rescale(originSG_ppp, 1000, "km")
destSG_ppp.km <- rescale(destSG_ppp, 1000, "km")4.2 Deriving Traditional Kernel Density Estimation (KDE) Layers
4.2.1 Automatic bandwidth selection method
We will first compute the kernel density by using density() of the spatstat package, with the default method bw.diggle().
Show code
kde_originSG_bw <- density(originSG_ppp.km,
sigma=bw.ppl,
edge=TRUE,
kernel="gaussian")
plot(kde_originSG_bw, main = "Kernel Density Estimation Layer")
Looking at all the different methods, we can see that bw.diggle() is still the best among the automatic bandwidth selection method.
Show code
bw.CvL(originSG_ppp.km) sigma
3.147573 Show code
bw.scott(originSG_ppp.km) sigma.x sigma.y
1.5973830 0.9321636 Show code
bw.ppl(originSG_ppp.km) sigma
0.1238747 Show code
bw.diggle(originSG_ppp.km) sigma
0.008087202 Show code
kde_originSG_ppl <- density(originSG_ppp.km,
sigma=bw.ppl,
edge=TRUE,
kernel="gaussian")
par(mfrow=c(1,2))
plot(kde_originSG_bw, main = "bw.diggle")
plot(kde_originSG_ppl, main = "bw.ppl")
4.2.2 Computing KDE by using fixed bandwidth
Having tried automatic bandwidth selection method, let’s try computing KDE by using a fixed bandwidth defined by us. In our case, we will define a fixed bandwidth of 800m (or 0.8km).
Show code
kde_originSG_500 <- density(originSG_ppp.km, sigma=0.5, edge=TRUE, kernel="gaussian")
plot(kde_originSG_500)
4.2.3 Computing KDE by using adaptive bandwidth
Fixed bandwidth method, however, is very sensitive to highly skewed distribution of spatial point patterns over geographical units, for example urban versus rural. To overcome this, we can try using adaptive bandwidth instead.
Show code
kde_childcareSG_adaptive <- adaptive.density(originSG_ppp.km, method="kernel")
plot(kde_childcareSG_adaptive)
4.2.4 Method we are using
As the KDE layer using fixed bandwidth with gaussian kernel plots a graph that allows for meaningful analysis at a glance, we will be using that for the steps moving forward.
Show code
kde_originSG_500 <- density(originSG_ppp.km, sigma=0.5, edge=TRUE, kernel="gaussian")
plot(kde_originSG_500)
Show code
kde_destSG_500 <- density(destSG_ppp.km, sigma=0.5, edge=TRUE, kernel="gaussian")
plot(kde_destSG_500)
Show code
par(mfrow=c(1,2))
plot(kde_originSG_500, main = "Origin KDE Layer")
plot(kde_destSG_500, main = "Destination KDE layer")
4.3 Combining KDE layers
4.3.1 Converting KDE layers into grid object
In order for us to map the KDE layer of these points to our map, we first need to convert it into grid object.
Show code
gridded_kde_originSG_500 <- as.SpatialGridDataFrame.im(kde_originSG_500)
spplot(gridded_kde_originSG_500)
Show code
gridded_kde_destSG_500 <- as.SpatialGridDataFrame.im(kde_destSG_500)
spplot(gridded_kde_destSG_500)
4.3.2 Converting KDE layers into grid object
We will then convert the gridded kernel density objects into RasterLayer object by using raster() of raster package. As the RasterLayer object does not include CRS information, we will need to manually assign it to them as well.
Show code
kde_originSG_500_raster <- raster(gridded_kde_originSG_500)
projection(kde_originSG_500_raster) <- CRS("+init=EPSG:3414 +datum=WGS84 +units=km")
kde_originSG_500_rasterclass : RasterLayer
dimensions : 128, 128, 16384 (nrow, ncol, ncell)
resolution : 0.4162063, 0.2250614 (x, y)
extent : 2.667538, 55.94194, 21.44847, 50.25633 (xmin, xmax, ymin, ymax)
crs : +proj=tmerc +lat_0=1.36666666666667 +lon_0=103.833333333333 +k=1 +x_0=28001.642 +y_0=38744.572 +datum=WGS84 +units=km +no_defs
source : memory
names : v
values : -1.552973e-14, 591.2389 (min, max)Show code
kde_destSG_500_raster <- raster(gridded_kde_destSG_500)
projection(kde_destSG_500_raster) <- CRS("+init=EPSG:3414 +datum=WGS84 +units=km")
kde_destSG_500_rasterclass : RasterLayer
dimensions : 128, 128, 16384 (nrow, ncol, ncell)
resolution : 0.4162063, 0.2250614 (x, y)
extent : 2.667538, 55.94194, 21.44847, 50.25633 (xmin, xmax, ymin, ymax)
crs : +proj=tmerc +lat_0=1.36666666666667 +lon_0=103.833333333333 +k=1 +x_0=28001.642 +y_0=38744.572 +datum=WGS84 +units=km +no_defs
source : memory
names : v
values : -1.361913e-14, 526.6182 (min, max)4.3.3 Overlaying KDE layer on tmap plot
To further explore the map, we will now be overlaying the KDE layer both onto OpenStreetMap of Singapore, and also on the Singapore Planning Area layer and OSM road layer that we have pre-processed.
4.3.3.1 Overlay on OpenStreetMap
Show code
tmap_mode("view")
tm_basemap(leaflet::providers$OpenStreetMap) +
tm_shape(kde_originSG_500_raster) +
tm_raster("v", alpha = 0.7,
palette = "YlOrRd") +
tm_layout(legend.position = c("right", "bottom"), frame = FALSE)Show code
tmap_mode("plot")Show code
tmap_mode("view")
tm_basemap(leaflet::providers$OpenStreetMap) +
tm_shape(kde_destSG_500_raster) +
tm_raster("v", alpha = 0.7,
palette = "YlOrRd") +
tm_layout(legend.position = c("right", "bottom"), frame = FALSE)Show code
tmap_mode("plot")As you can see from the plot, there are certain planning areas that are hotspots for hailing of Grab ride service, in particular Central Region (Orchard, Newton etc), Woodlands, Punggol, Tampines, and most notably Changi (where the airport lies).
4.3.3.2 Overlay on Planning Area and OSM Road Layers
To further confirm our observation, let’s plot the KDE layer over our Planning Area and OSM Road Layers.
Show code
tmap_mode("view")
tm_shape(mpsz2019_new) +
tm_polygons("PLN_AREA_N") +
tm_shape(kde_originSG_500_raster) +
tm_raster("v", alpha = 0.7,
palette = "YlOrRd")Show code
tmap_mode("view")
tm_shape(mpsz2019_new) +
tm_polygons("PLN_AREA_N") +
tm_shape(kde_destSG_500_raster) +
tm_raster("v", alpha = 0.7,
palette = "YlOrRd")The common overlapping Planning Areas include “TAMPINES”, “CHANGI”, “WOODLANDS”, and “NOVENA”, so let’s do a further analysis on these areas.
4.4 In-depth KDE Computation
4.4.1 Data Preparation
To do in-depth KDE computation on these 4 planning areas, we will first need to extract their respective boundaries. In the code below, we extracted their boundaries and converted them to sp’s Spatial* class.
Show code
mpsz <- as_Spatial(mpsz2019_new)
cg = mpsz[mpsz@data$PLN_AREA_N == "CHANGI",]
tp = mpsz[mpsz@data$PLN_AREA_N == "TAMPINES",]
wl = mpsz[mpsz@data$PLN_AREA_N == "WOODLANDS",]
nv = mpsz[mpsz@data$PLN_AREA_N == "NOVENA",]Plotting down these boundaries.
Show code
par(mfrow=c(2,2))
plot(cg, main = "CHANGI")
plot(tp, main = "TAMPINES")
plot(wl, main = "WOODLANDS")
plot(nv, main = "NOVENA")
Turning the spatial point data frame into generic sp format, then into owin object as done previously.
Show code
cg_sp = as(cg, "SpatialPolygons")
tp_sp = as(tp, "SpatialPolygons")
wl_sp = as(wl, "SpatialPolygons")
nv_sp = as(nv, "SpatialPolygons")
cg_owin = as(cg_sp, "owin")
tp_owin = as(tp_sp, "owin")
wl_owin = as(wl_sp, "owin")
nv_owin = as(nv_sp, "owin")By using the code below, we will be able to extract grab origin and destination points for these specific areas.
Show code
origin_cg_ppp = origin_ppp[cg_owin]
origin_tp_ppp = origin_ppp[tp_owin]
origin_wl_ppp = origin_ppp[wl_owin]
origin_nv_ppp = origin_ppp[nv_owin]
dest_cg_ppp = dest_ppp[cg_owin]
dest_tp_ppp = dest_ppp[tp_owin]
dest_wl_ppp = dest_ppp[wl_owin]
dest_nv_ppp = dest_ppp[nv_owin]Next up is the rescale() function used previously as well.
Show code
origin_cg_ppp.km = rescale(origin_cg_ppp, 1000, "km")
origin_tp_ppp.km = rescale(origin_tp_ppp, 1000, "km")
origin_wl_ppp.km = rescale(origin_wl_ppp, 1000, "km")
origin_nv_ppp.km = rescale(origin_nv_ppp, 1000, "km")
dest_cg_ppp.km = rescale(dest_cg_ppp, 1000, "km")
dest_tp_ppp.km = rescale(dest_tp_ppp, 1000, "km")
dest_wl_ppp.km = rescale(dest_wl_ppp, 1000, "km")
dest_nv_ppp.km = rescale(dest_nv_ppp, 1000, "km")Finally, we plot the four planning areas and the grab hailing origin and destination points
Show code
par(mfrow=c(2,4))
plot(origin_cg_ppp.km, main = "CHANGI ORIGIN")
plot(origin_tp_ppp.km, main = "TAMPINES ORIGIN")
plot(origin_wl_ppp.km, main = "WOODLANDS ORIGIN")
plot(origin_nv_ppp.km, main = "NOVENA ORIGIN")
plot(dest_cg_ppp.km, main = "CHANGI DESTINATION")
plot(dest_tp_ppp.km, main = "TAMPINES DESTINATION")
plot(dest_wl_ppp.km, main = "WOODLANDS DESTINATION")
plot(dest_nv_ppp.km, main = "NOVENA DESTINATION")
4.4.2 Computing KDE
We will now be computing the KDE of each planning area using the fixed bandwidth method.
Show code
par(mfrow=c(1,2))
plot(density(origin_cg_ppp.km,
sigma=0.5,
edge=TRUE,
kernel="gaussian"),
main="Changi Origin")
plot(density(dest_cg_ppp.km,
sigma=0.5,
edge=TRUE,
kernel="gaussian"),
main="Changi Destination")
Show code
tmap_mode('plot')
tm_shape(mpsz2019_new[mpsz2019_new$PLN_AREA_N == "CHANGI", ]) +
tm_polygons('SUBZONE_N')
The hotspot in Changi area is centered around Changi Airport, indicating a likely surge in use of Grab services due to the constant flow of passengers arriving and departing from Singapore.
Show code
par(mfrow=c(1,2))
plot(density(origin_tp_ppp.km,
sigma=0.5,
edge=TRUE,
kernel="gaussian"),
main="Tampines Origin")
plot(density(dest_tp_ppp.km,
sigma=0.5,
edge=TRUE,
kernel="gaussian"),
main="Tampines Destination")
Show code
tmap_mode('plot')
tm_shape(mpsz2019_new[mpsz2019_new$PLN_AREA_N == "TAMPINES", ]) +
tm_polygons('SUBZONE_N')
The hotspot in Tampines area is mainly concentrated around the stretch from Tampines West to Tampines East, encompassing the bulk of where most residents of Tampines currently live (Tampines West, Tampines, Tampines East).
Show code
par(mfrow=c(1,2))
plot(density(origin_wl_ppp.km,
sigma=0.5,
edge=TRUE,
kernel="gaussian"),
main="Woodlands Origin")
plot(density(dest_wl_ppp.km,
sigma=0.5,
edge=TRUE,
kernel="gaussian"),
main="Woodlands Destination")
Show code
tmap_mode('plot')
tm_shape(mpsz2019_new[mpsz2019_new$PLN_AREA_N == "WOODLANDS", ]) +
tm_polygons('SUBZONE_N')
The rides are concentrated around the lower half of Woodlands area, ranging from Woodlands West to Woodlands South, then Woodlands East. However, one prominent hotspot shared across both the origin and destination map is the Woodlands West region, indicating that this might either be the area with the wealthiest residents in Woodlands, or that there are just more residents concentrated here.
Show code
par(mfrow=c(1,2))
plot(density(origin_nv_ppp.km,
sigma=0.5,
edge=TRUE,
kernel="gaussian"),
main="Novena Origin")
plot(density(dest_nv_ppp.km,
sigma=0.5,
edge=TRUE,
kernel="gaussian"),
main="Novena Destination")
Show code
tmap_mode('plot')
tm_shape(mpsz2019_new[mpsz2019_new$PLN_AREA_N == "NOVENA", ]) +
tm_polygons('SUBZONE_N')
The Novena area’s notable hotspots present an interesting distinction. Origin points predominantly converge around the affluent Moulmein area, revealing a concentration in the wealthier section of town. Conversely, the destination points gravitate towards the Malcolm area, characterized by a cluster of prestigious schools, as illustrated in the figure below.
4.5 Nearest Neighbour Analysis
In this section, we will perform the Clark-Evans test of aggregation for a spatial point pattern by using clarkevans.test() of statspat package, to test whether the distribution of Grab ride hailing origin points are randomly distributed.
Using 95% confidence interval, the test hypotheses are:
Ho = The distribution of Grab ride hailing origin points are randomly distributed.
H1= The distribution of Grab ride hailing origin points are not randomly distributed.
For this section, we will be making use of the ppp object.
Clark-Evans Test
Show code
clarkevans.test(originSG_ppp,
correction="none",
clipregion="sg_owin",
alternative=c("clustered"),
nsim=99)
Clark-Evans test
No edge correction
Z-test
data: originSG_ppp
R = 0.27981, p-value < 2.2e-16
alternative hypothesis: clustered (R < 1)Show code
clarkevans.test(origin_cg_ppp,
correction="none",
clipregion="sg_owin",
alternative=c("clustered"),
nsim=99)
Clark-Evans test
No edge correction
Z-test
data: origin_cg_ppp
R = 0.11405, p-value < 2.2e-16
alternative hypothesis: clustered (R < 1)Show code
clarkevans.test(origin_tp_ppp,
correction="none",
clipregion="sg_owin",
alternative=c("clustered"),
nsim=99)
Clark-Evans test
No edge correction
Z-test
data: origin_tp_ppp
R = 0.32408, p-value < 2.2e-16
alternative hypothesis: clustered (R < 1)Show code
clarkevans.test(origin_wl_ppp,
correction="none",
clipregion="sg_owin",
alternative=c("clustered"),
nsim=99)
Clark-Evans test
No edge correction
Z-test
data: origin_wl_ppp
R = 0.31779, p-value < 2.2e-16
alternative hypothesis: clustered (R < 1)Show code
clarkevans.test(origin_nv_ppp,
correction="none",
clipregion="sg_owin",
alternative=c("clustered"),
nsim=99)
Clark-Evans test
No edge correction
Z-test
data: origin_nv_ppp
R = 0.34615, p-value < 2.2e-16
alternative hypothesis: clustered (R < 1)Having performed the Clark-Evans Test on all 4 planning area and Singapore as a whole, all of their p-values are <2.2e-16 < 0.05, thus we reject Ho. This means that the distribution of Grab ride hailing origin points are not randomly distributed which we explored in earlier sections.
Furthermore, as their R value ranges from 0.11647 to 0.35838 which is <1, this suggests that the points are clustering.
5. Network Kernel Density Estimation (NKDE) Layers
In this section, we will be using appropriate functions of spNetwork package:
- to derive network constrained kernel density estimation (NetKDE), and
- to perform network G-function and k-function analysis,
where in this case the network refers to OSM’s Road Map of Singapore.
However, due to limitations in computational power, we will be limiting the area of scope down to the 4 areas identified in the previous section, Changi, Tampines, Woodlands, and Novena, and only the Origin points.
5.1 Data Preparation
5.1.1 Initial Data Pre-processing
Before we begin, let us first convert our sg_roads_new data from SFC_GEOMETRY to SFC_LINESTRING.
Show code
sg_roads_linestring <- st_cast(sg_roads_new, "LINESTRING")5.1.2 Narrowing down the scope
Then, let us narrow down the scope of our data to the 4 areas mentioned.
Show code
# Roads
cg_roads <- sg_roads_linestring %>% filter(PLN_AREA_N == "CHANGI")
tp_roads <- sg_roads_linestring %>% filter(PLN_AREA_N == "TAMPINES")
wl_roads <- sg_roads_linestring %>% filter(PLN_AREA_N == "WOODLANDS")
nv_roads <- sg_roads_linestring %>% filter(PLN_AREA_N == "NOVENA")
# Grab Origin Points
cg_origin <- origin_sf_new %>% filter(PLN_AREA_N == "CHANGI")
tp_origin <- origin_sf_new %>% filter(PLN_AREA_N == "TAMPINES")
wl_origin <- origin_sf_new %>% filter(PLN_AREA_N == "WOODLANDS")
nv_origin <- origin_sf_new %>% filter(PLN_AREA_N == "NOVENA")5.1.3 Visualising the data
Before we begin our analysis, let us visualise our geospatial data to make sure everything falls into place.
Show code
tm_shape(cg_roads) +
tm_lines("PLN_AREA_N") +
tm_shape(cg_origin) +
tm_dots()
Show code
tm_shape(tp_roads) +
tm_lines("PLN_AREA_N") +
tm_shape(tp_origin) +
tm_dots()
Show code
tm_shape(wl_roads) +
tm_lines("PLN_AREA_N") +
tm_shape(wl_origin) +
tm_dots()
Show code
tm_shape(nv_roads) +
tm_lines("PLN_AREA_N") +
tm_shape(nv_origin) +
tm_dots()
5.2 Network Constrained KDE (NetKDE) Analysis
We will now perform NetKDE analysis by using appropriate functions provided in spNetwork package.
5.2.1 Preparing the lixels objects
Before we can compute NetKDE, the SpatialLines object need to be cut into lixels with a specified minimal distance, and this can be done using lixelize_lines() of spNetwork package.
Show code
cg_lixels <- lixelize_lines(cg_roads,
700,
mindist = 350)
tp_lixels <- lixelize_lines(tp_roads,
700,
mindist = 350)
wl_lixels <- lixelize_lines(wl_roads,
700,
mindist = 350)
nv_lixels <- lixelize_lines(nv_roads,
700,
mindist = 350)5.2.2 Generating line centre points
Next, we will use lines_center() of spNetwork to generate a SpatialPointsDataFrame (i.e. samples) with line centre points.
Show code
cg_lines_center <- lines_center(cg_lixels)
tp_lines_center <- lines_center(tp_lixels)
wl_lines_center <- lines_center(wl_lixels)
nv_lines_center <- lines_center(nv_lixels)5.2.3 Computing NetKDE
We are now ready to compute NetKDE. As the code is fairly long, we will split it into 4 tabs.
Show code
# Origin
cg_o_densities <- nkde(cg_roads,
events = cg_origin,
w = rep(1,nrow(cg_origin)),
samples = cg_lines_center,
kernel_name = "quartic", # kernel method
bw = 300,
div= "bw",
method = "simple",
# method used to calculate NKDE. spNetwork supports 3 popular methods, namely simple, discontinuous, and continuous
digits = 1,
tol = 1,
grid_shape = c(1,1),
max_depth = 8,
agg = 5,
# we aggregate events within a 5m radius (faster calculation)
sparse = TRUE,
verbose = FALSE)Show code
tp_o_densities <- nkde(tp_roads,
events = tp_origin,
w = rep(1,nrow(tp_origin)),
samples = tp_lines_center,
kernel_name = "quartic", # kernel method
bw = 300,
div= "bw",
method = "simple",
# method used to calculate NKDE. spNetwork supports 3 popular methods, namely simple, discontinuous, and continuous
digits = 1,
tol = 1,
grid_shape = c(1,1),
max_depth = 8,
agg = 5,
# we aggregate events within a 5m radius (faster calculation)
sparse = TRUE,
verbose = FALSE)Show code
wl_o_densities <- nkde(wl_roads,
events = wl_origin,
w = rep(1,nrow(wl_origin)),
samples = wl_lines_center,
kernel_name = "quartic", # kernel method
bw = 300,
div= "bw",
method = "simple",
# method used to calculate NKDE. spNetwork supports 3 popular methods, namely simple, discontinuous, and continuous
digits = 1,
tol = 1,
grid_shape = c(1,1),
max_depth = 8,
agg = 5,
# we aggregate events within a 5m radius (faster calculation)
sparse = TRUE,
verbose = FALSE)Show code
nv_o_densities <- nkde(nv_roads,
events = nv_origin,
w = rep(1,nrow(nv_origin)),
samples = nv_lines_center,
kernel_name = "quartic", # kernel method
bw = 300,
div= "bw",
method = "simple",
# method used to calculate NKDE. spNetwork supports 3 popular methods, namely simple, discontinuous, and continuous
digits = 1,
tol = 1,
grid_shape = c(1,1),
max_depth = 8,
agg = 5,
# we aggregate events within a 5m radius (faster calculation)
sparse = TRUE,
verbose = FALSE)5.2.4 Reinsert Density
Before we are able to visualise, we first need to insert the computed values back into lines_center and lixels objects as density field.
Show code
cg_lines_center$o_density <- cg_o_densities
cg_lixels$o_density <- cg_o_densities
tp_lines_center$o_density <- tp_o_densities
tp_lixels$o_density <- tp_o_densities
wl_lines_center$o_density <- wl_o_densities
wl_lixels$o_density <- wl_o_densities
nv_lines_center$o_density <- nv_o_densities
nv_lixels$o_density <- nv_o_densitiesSince svy21 projection system is in meter, the computed density values are very small i.e. 0.0000005. We will thus need to rescale the density values from number of events per meter to number of events per kilometer.
Show code
cg_lines_center$o_density <- cg_lines_center$o_density*1000
cg_lixels$o_density <- cg_lixels$o_density*1000
tp_lines_center$o_density <- tp_lines_center$o_density*1000
tp_lixels$o_density <- tp_lixels$o_density*1000
wl_lines_center$o_density <- wl_lines_center$o_density*1000
wl_lixels$o_density <- wl_lixels$o_density*1000
nv_lines_center$o_density <- nv_lines_center$o_density*1000
nv_lixels$o_density <- nv_lixels$o_density*10005.2.5 Visualising NetKDE
Show code
tmap_mode('view')
tm_basemap(leaflet::providers$OpenStreetMap) +
tm_shape(cg_lixels)+
tm_lines(col="o_density")+
tm_shape(cg_origin)+
tm_dots(alpha=0.2)Show code
tmap_mode('plot')This tmap plot further reinforces our observation above that the grab ride traffic are from incoming tourists or locals returning home form the airport, as you can see the denser area being the Changi Airport Terminals. However, it is worth highlighting that there some slight traffic along the Changi Village area and infront of the Japanese School as well.
Show code
tmap_mode('view')
tm_basemap(leaflet::providers$OpenStreetMap) +
tm_shape(tp_lixels)+
tm_lines(col="o_density")+
tm_shape(tp_origin)+
tm_dots(alpha=0.2)